Back to Models

scTour

🚀 Trajectory Inference

Trajectory Inference and Ordering

RNA
ATAC

Trajectory inference VAE for learning cell developmental paths in single-cell data

Publications

Deep generative modeling of transcriptional dynamics for RNA velocity analysis in single cells

Li et al.2023
Complexity
★★☆
moderate
Interpretability
★★★
high
Architecture
Trajectory VAE
Latent Dim
10

Latent Trajectory Inference

scTour learns latent developmental trajectories by modeling cell progression through ordered latent representations with VAE reconstruction

Main Idea

Infer cell differentiation trajectories by learning smooth paths through latent space with reconstruction

Key Components

Trajectory Encoder

VAE encoder with ordered latent structure

Path Smoothness

Regularization for smooth trajectories

Pseudo-timing

Assigns pseudo-time along trajectories

Decoder

Reconstructs expression from trajectory embeddings

Mathematical Formulation

z_t = f_θ(t); smooth ordering via latent trajectory; X̂ = Decoder(z)

Loss Functions

ELBO
Reconstruction + KL
Smoothness
Temporal regularization

Data Flow

Expression → Encoder → Ordered Latent → Trajectory Inference → Decoder → Pseudo-time + Reconstructed Expression

Architecture Details

Architecture Type

VAE with Temporal Structure

Input/Output Types

single-cell → reconstruction

Key Layers

EncoderTemporalDecoderTrajectoryHead

Frameworks

PyTorch

Tags

trajectorypseudo-timedevelopmentvaegenerativerna